Failed to route the following 1 signal(s)-- global signals

By the looks of the number of clocks you have are stressing the global routing resources of the design.

The simplest for me to say (but hardest to implement) is to simply reduce the number of clocks in your design.

It appears a difference between "Working" and "non-working" layouts is the number of regional clocks. Specifically, it appears the TX_DAC clocks gets split into two regions, versus one.

I would start by locking down specific logic to specific regions in the design. (Especially associated with the regional clocks, or smaller clock domains). This should allow the results to be more repeatable at least.

Pete